To tackle the problems of insufficiently extracting polarimetric information from PolSAR image and low classification accuracy of H/Alpha/A-Wishart unsupervised classification algorithm, this paper proposes an adapted algorithm named MCSM-Wishart by imposing multiple-component scattering model (MCSM)decomposition to fit unsupervised classification of polarimetric SAR image. Firstly, various kinds of polarimetric information such as volume scatter, double scatter, helix scatter, surface scatter and wire scatter can be extracted from the image by MCSM decomposition, and iterative self-organizing data analysis(ISODATA)technique is used for clustering. Then iterative classification based on complex Wishart distribution is used to obtain the final result. H/Alpha-Wishart, H/Alpha/A-Wishart, MCSM-Wishart and supervised-Wishart algorithms are compared with each other based on two research plots conducted respectively in Lishui of Nanjing City and Binhai Wetland of Yancheng City with PALSAR image from ALOS. The results show that MCSM-Wishart classification algorithm can improve to a certain extent the original classifiers in terms of efficiency, total accuracy and Kappa coefficient. It is therefore concluded that the polarimetric information extracted by MCSM decomposition can sufficiently reflect the characteristics of the ground object. Combining with ISODATA clustering algorithm, MCSM decomposition can be used in the iterative classification based on complex Wishart distribution so as to improve the classification accuracy and reliability efficiently.
Atmospheric correction is the basic step in quantitative retrieval of land surface parameters with hyperspectral imagery. Based on abundant spectral information in the hyperspectral image,this paper presents a new atmospheric correction algorithm for hyperspectral imagery characterized by collaborative retrieval of the aerosol optical thickness (AOT) and the water vapor content (WV). The algorithm takes into account the effects of aerosol type,AOT and WV,and uses the iteration method combined with the 6S(second simulation of the satellite signal in the solar spectrum)radiative transfer model to retrieve the atmospheric parameters and ground reflectance. This new method overcomes the weakness of the existing atmospheric correction algorithms which fail to consider the effects of both AOT and WV. Hyperion hyperspectral image data covering Wuhan City were used to verify the effectiveness of the algorithm proposed in this paper,with the results compared with those of FLAASH(fast line-of-sight atmospheric analysis of spectral hypercubes)method in ENVI and MODIS's AOT and WV products. It is shown that the proposed algorithm can better correct the effect of aerosol and water vapor in the atmosphere,and needs no additional parameters because all the inputs are taken from the image data themselves or the 6S radiative transfer model in the inversion process.
A new strategy for the classification of raw LiDAR points in urban areas, which is based on the comprehensive utilization of echo features of different object types and terrain information, is proposed in this paper according to a regional multi-return density analysis. The main procedure of the classification of the off-terrain points begins with the construction of Triangulated Irregular Network (TIN), and then the region of each object is captured by the contours clustering based on the topological relations of various contours traced from the TIN. Finally, the type of the object is recognized by the statistical analysis of the regional multi-return density through the significant difference between the building region and the vegetation region. This method not only makes good use of the difference in echo features between different objects such as buildings and trees but also confirms the existence of the multi-returns on the edges of the building. At the same time, the adaptive region determination of the objects is accomplished following the contours clustering. So the proposed method can dramatically increase the classification accuracy and overcome the weakness of the traditional methods, thus being more useful to the study and application of such aspects as building reconstruction and parameters estimation of the trees. Experiments prove that the new algorithm can get an effective classification.
The topographic correction is the most critical component part of the remote sensing quantitative study of rugged terrain areas. According to the idea of slope grading in combination with the simplified Three Factor + C model, a Three Factor +C +Slope model was established to eliminate the defect of traditional and empirical topographic correction using the same coefficient as the slope changing. The results show that the Three Factor+C+ Slope model is better than the C model, the SCS model, the Three Factor model and the Three Factor + C model in six calibration test indicators comprising the mean value, the standard deviation, the correlation between pixel value and illumination coefficient, the radiance discrepancy before and after correction, the dispersion index and the homogeneity coefficient. Due to its advantages such as excellent physical mechanism and considerable removal of the terrain effects on radiance, the Three Factor + C + Slope model is feasible and worthy of promotion.
In this paper, an improved method for multi-spectral and panchromatic remote sensing image fusion was proposed. The method proposed was inspired by the traditional wavelet and IHS based fusion framework. The Contourlet transformation was utilized and a novel fusion strategy was presented, with the purpose of extracting the luminance component of the multispectral image and conducting the Contourlet transformation on the luminance component and the panchromatic images. The fusion strategy for the low-frequency component is to adjust adaptively, while the high-frequency is threshold-controlled based on the structure similarity. Extensive experiments show that the method proposed in this paper can effectively extract the spatial information from the panchromatic image, which is not present in the multi-spectral image. The quantitative evaluation results also suggest that the resultant image has a higher correlation coefficient with the original image and smaller spectral distortion degrees in comparison with images of the conventional methods. In addition, the information entropy and the standard deviations are also superior. Besides, the proposed method is to a certain extent practical.
The detection accuracy is likely to be influenced by water bodies and bluish surface features during the detection of shadows on high-resolution remote sensing images. To tackle this problem, this paper proposes a new shadow detection method using principal component transform and multi-band operation. Firstly, the spectral values of typical surface features such as shadows, water bodies and buildings are counted and analyzed in QuickBird images. Secondly, the non-shaded area and shaded area are identified based on principal component transform combined multi-band operation and automatically detected by multimodal histogram threshold algorithm. Finally, the detected result is processed by morphological filtering algorithm. The result shows that this method shows higher extraction accuracy, efficiency and universality for QuickBird images.
Because the choice of the reasonable and effective threshold of segmentation is rather difficult, the method based on threshold segmentation is not applicable to extracting road information from remote sensing images in that there are obviously multiple types of road and non-road feature interference. To tackle this problem, the authors propose in this paper a method which combines Mean Shift algorithm with threshold segmentation to extract road information. Firstly, the Mean Shift is used to smooth the image, then the texture distribution is made more uniform in the road and the edge of the road is kept. Secondly, the Mean Shift segmentation is used to segment the image, and group the roads which have the same or similar gray values into one gray value showing. Thirdly, as different kinds of color information of the road have different showing characteristics in the gray-level histogram, the gray value which has relatively more numbers of picture elements is taken as the segment range boundary point to obtain the original road information by using multi-threshold segmentation. Finally, post-processing of the original road information is made to obtain the road. The experiment results indicate that this method can extract the road information from the remote sensing imagery and broaden the scope of the use of the threshold segmentation to extract the road information.
Forest vegetation remote sensing image segmentation is an important kind of target, and effective determination of the scale of forest vegetation texture segmentation is an important research topic. This paper presents a method in which the blue noise theory is used to describe the characteristics of remote sensing images for forest vegetation texture. This is a new method for vegetation texture characterization and texture scale calculation. The correspondence between the research scale morphology and vegetation textures can be used in the selected detection area to iteratively search for blue noise characteristics. Iteration consists of the reduction of the region size through the geometric transformation and the obtaining of a spectral response region by fast Fourier transform so as to extract the blue noise characteristics from the spectral response. For regions with blue noise characteristics, the intensity distribution of forest vegetation texture is computed, and the texture size is calculated based on the current size of the area. Experimental results show that the gray scale and the distribution of forest vegetation texture units can be accurately measured, which lays reliable foundation for further texture segmentation.
Leaf area index(LAI)is an important parameter for decrypting canopy structure, and the accurate acquisition of orchards LAI plays an important role in monitoring the growing condition and estimating the yield of orchards. In this paper, the orchard blocks in the middle of California in USA were chosen as the study area, and LAI was retrieved using normalized difference water index (NDWI) through comparing regression models with surveyed leaf area indices and three vegetation indices composed of normalized difference vegetation index (NDVI), normalized difference infrared index (NDII) and NDWI based on the MODIS/ASTER airborne simulator(MASTER)image, which acquired flying along the solar plane. The results show that the image acquired by flying perpendicular to the solar plane has the phenomenon of maximum brightness gradients because of the bidirectional reflectance of surface object, whereas the image acquired by flying along the solar plane fails to show such a phenomenon. The comparison between three vegetation index models also shows that NDVI is easy to reach saturation in higher coverage area, and NDWI is more suitable for LAI retrieval in the study area because NDWI model has higher R2 and smaller RMSE. The results of this study can enrich the LAI retrieval theory and provide theoretical and data support for LAI scale problem.
This paper introduces level set theory to the feature extraction of coastline contour information. In this paper, the author first reviewed the related research work in this field and describes the level set theory and its applications, and then proposed the coastline contour segmentation algorithm and area smooth nonparametric density estimation before using it to extract different kinds of coastlines. To illustrate the effectiveness of the level set method (LSM) algorithm in coastline feature extracting, this paper compared the LSM algorithm and gradient descent method to demonstrate the coastline feature extraction efficiency of LSM. The optical and remote-sensing images used in experimental tests were of different contour features, multi-resolution and different point of views. The results achieved show that the level set algorithm is robust in analyzing characteristics of the coast complex texture even with the influence of noise. Also, it has strong sensitivity in edge information detection and is capable of quickly and effectively extracting features from the boundary information.
Satellite image simulation technology aims at simulating the band features, space geometric features, radiation characteristics, ephemeris data and format of the satellite image by the method of computer simulation before the launching of satellites. To study the Landsat image simulation technology,this paper makes a review on the development of China's own satellite remote sensing image simulation system over the past six years,and describes the design of China's image simulation system as well as the key technology. At present, the simulated bands of the system include the bands from visible light, near infrared to thermal infrared band, with the simulated spatial resolution between 3 m to 300 m. In the process of simulation, the authors used the remote sensing radiative transfer model to simulate the spectrum characteristics, employed the PROSPECT+SAIL models to simulate the spectrum of the areas covered by vegetation, and adopted spectral library to simulate the spectrum of the non-vegetation area. Based on linear decomposition of the atmospheric radiative transfer process, the authors set up a look-up table (LUT)of the atmospheric radiative transfer process so as to improve the speed of simulation calculation on the premise of guaranteeing simulation precision significantly. In order to simulate the precise geometry information, the authors used high precision geometric positioning model on the basis of considering the topographic relief, calculated the intersection between the line of sight for satellite observation and the Earth's surface pixel by pixel. Finally, The authors used the observation data after the launching of the satellite and the field measured data in the experimental field to verify the simulation precision of the image simulation technology described in this paper.
Spaceborne wide-field imaging spectrometer sets up two channels, CH18(8.125~8.825 μm) and CH19 (8.925~9.275 μm) for retrieving land surface temperature. To verify the adaptability of the traditional split-window algorithms applied to the atmospheric window of 10~14 μm for the atmospheric window of 8.0~9.3 μm, the authors introduced three kinds of split-window algorithms, i.e., Sobrino,Franca & Cracknell and Becker, with the parameter calculation formulae being revised against band setting for split-window in atmospheric window of 8.0~9.3 μm, and verified the retrieved accuracies of land surface temperature by the six standard atmospheric models supplied by MODTRAN. The results show that the present 3 kinds of split-window algorithms fail to meet the land surface temperature precision requirement of less than 1K and are not suitable for direct transplantation.
To analysis the relationship between the hyperspectral reflectance in the visible/near infrared bands and available nitrogen (AN) in paddy soil in southern China hilly areas, the authors collected the hyperspectral reflectance of paddy soil and made analysis with spectral analysis methods with the purpose of discovering the spectral characteristics of field reflectance and its influencing factors. The spectral indices were derived, and then paddy soil AN predicting model based on the correlation between AN content and spectral indices was built. The results were as follows:The different AN content paddy soil reflectance curves showed the tendency that, with the increase of AN content, the spectral reflectance decreased and the absorption depth became greater;by analyzing the correlation coefficient of paddy soil AN content and 16 kinds of mathematical transformations of spectral reflectance, the sensitive wavelengths were extracted, which were 694 nm, 2 058 nm and 2 189 nm; the predicting model for paddy soil AN content was built with spectral resample reflectance at 694 nm, 2 058 nm and 2 189 nm as independent variables and AN as dependent variable, and the coefficients of determination R2 of the model was 0.56, suggesting that the model is quite good in stability and predictability.
To establish the remote sensing monitoring model for soil salinization, the authors chose the typical soil salinization area in Pingluo County of Ningxia as the study area, and measured the spectral data in the field. These data, together with the values of pH and salinity measured in the laboratory, were taken as the basic data. Hyperspectral data processing method was used to analyze the spectral characteristics of different levels of soil salinization. Spectral data were transformed with 11 different approaches, such as reciprocal, logarithm, root mean square and their first order differentials. After the transformation, the correlation analysis was carried out between the obtained soil spectra and soil salinity. The most sensitive band was selected, and the field spectral sensitive band and soil salinity were used and the multiple linear regression was employed to establish the spectral quantitative models for evaluating the soil salinization degrees. The results show that the reciprocal first order differential of measured soil spectral is most sensitive to soil salinization degrees. The spectral quantitative models based on the wavelengths of 940 nm and 1 094 nm are the best.